John Darrington [Fri, 1 Jul 2011 16:20:42 +0000 (18:20 +0200)]
Fix bugs deleting temporary files and their containing directory.
The clean_temp module provided by gnulib doesn't cleanup on program
exit. We must do that ourselves. Secondly, the close_temp_file function
was supposed to remove the file, but didn't do that. Thirdly, files
were not being registered with the clean_temp module on creation.
This change fixes these problems.
John Darrington [Fri, 1 Jul 2011 16:18:27 +0000 (18:18 +0200)]
Change the order of the reported NPAR values in two sample related tests.
This change makes the order of the reported values follow the order in
which they were entered in the syntax. The changes to the tests affect
only the order. The numbers are unchanged.
John Darrington [Fri, 1 Jul 2011 14:47:33 +0000 (16:47 +0200)]
Re-implemented the T-TEST command and the levene calculation.
Also updated the ONEWAY command to use the new levene test.
The changes to t-test.at in this changeset affect only whitespace
or the order of reported significance. The numbers remain unchanged.
John Darrington [Mon, 27 Jun 2011 07:13:11 +0000 (09:13 +0200)]
GLM: Add unimplemented subcommands, and add a test.
This change adds a basic GLM test for the "Latin Square" use case.
It also implements the more common subcommands that might occur in
a real life example of such usage. The /DESIGN subcommand differs
slightly from spss. In spss, it is optional, and if given can be
empty. In this change, /DESIGN is mandatory and may not be empty.
The rational for that, is that spss's default is to include interactions
for all the factors. We don't (yet) support interactions, so it's
better to refuse to run a syntax which relies upon the default
rather than to run it with different semantics.
Ben Pfaff [Sat, 11 Jun 2011 17:40:43 +0000 (10:40 -0700)]
gui: Eliminate some dependencies on the dictionary.
Previously all of the dialog boxes used by the variable sheet
kept around a pointer to the dictionary solely to obtain the
dictionary's encoding. Previous commits have eliminated that
use, so this commit gets rid of the dictionary members.
Ben Pfaff [Tue, 21 Jun 2011 02:06:11 +0000 (19:06 -0700)]
gui: Simplify value_to_text() parameters.
Both the format and the dictionary can be inferred from the
variable being formatted in every case in the tree but one, so
use that as the calling convention. In the one remaining case,
it's just as easy to call data_out() directly.
Ben Pfaff [Thu, 23 Jun 2011 05:13:40 +0000 (22:13 -0700)]
sys-file-reader: Reduce invalid format from error to warning.
gaby jongenelen <geebie21@gmail.com> provided a .sav file
written by "MS Windows Release 13.0 spssio32.dll" that had
zeros in the "write" format field for several numeric
variables. Until now, PSPP treated this as an error and
rejected the file. This commit changes the error to a warning
and suppresses the diagnostic entirely for this case of a
zero value.
Ben Pfaff [Sat, 18 Jun 2011 05:21:38 +0000 (22:21 -0700)]
sys-file-writer: Write identical sets of value labels only once.
The .sav file format can save space by writing a single set of
value labels once and assigning it to multiple variables. PSPP
has never taken advantage of this, instead writing the value
labels for each variable separately. This commit adopts the
strategy for avoiding duplicates apparently used by SPSS, based
on observing a .sav file that had many groups of variables with
the same labels.
Thanks to Curt Reinhold <Curt.Reinhold@psychonomics.de> for
providing the .sav file that led to this discovery.
Ben Pfaff [Sat, 18 Jun 2011 05:18:27 +0000 (22:18 -0700)]
variable: Do not trim trailing whitespace from variable labels.
Until now, PSPP has trimmed trailing whitespace from variable
labels. SPSS does not do this, which led to gratuitous changes
in variable labels when reading and then writing back a .sav
file, so this commit stops stripping trailing whitespace from
variable labels.
Thanks to Curt Reinhold <Curt.Reinhold@psychonomics.de> for
providing the .sav file that led to this discovery.
Ben Pfaff [Sat, 18 Jun 2011 05:16:44 +0000 (22:16 -0700)]
sys-file-writer: Omit empty multiple response sets records.
Multiple response sets are divided into two records in a .sav
file: one record with subtype 7 and one with subtype 19. PSPP
was incorrectly writing both records even if one of them was
empty. This commit makes it omit the empty record.
Thanks to Curt Reinhold <Curt.Reinhold@psychonomics.de> for
providing the .sav file that led to this discovery.
Ben Pfaff [Thu, 16 Jun 2011 06:13:12 +0000 (23:13 -0700)]
sys-file-reader, sys-file-writer: Use codepage numbers.
PSPP has had a library for converting between encoding names
and codepage numbers for a while, but the system file reader
and writer code has not taken advantage of it. This commit
make them use it.
Ben Pfaff [Thu, 16 Jun 2011 06:08:13 +0000 (23:08 -0700)]
Update code that assumed a dictionary's encoding could be NULL.
Until recently, I think, it was possible for dict_get_encoding()
to return NULL. It is no longer possible as far as I can see,
so this commit adds a comment that says so to the definition
of dict_get_encoding(). It also updates a few callers that
had checks for null pointers.
Ben Pfaff [Thu, 16 Jun 2011 01:55:54 +0000 (18:55 -0700)]
sys-file-writer: Put long variable name in variable attributes.
Otherwise, PSPP cannot read back .sav files that it writes, if
they contain variable attributes for variables with long names.
Thanks to Curt Reinhold <Curt.Reinhold@psychonomics.de> for
providing the .sav file that led to this discovery and to John
Darrington for helping to investigate.
Ben Pfaff [Wed, 15 Jun 2011 05:10:24 +0000 (22:10 -0700)]
GET DATA /TYPE=TXT: Also allow full format to be specified.
The implementation of fixed-format mode only allowed a format
type (e.g. DOLLAR) to be specified. SPSS also allowed a full
format specification (e.g. DOLLAR12.2), so this implements that
feature.
Reported-by: Ronaldo Baltar <ronaldo.baltar@gmail.com>
Ben Pfaff [Fri, 13 May 2011 05:10:07 +0000 (22:10 -0700)]
gui: Make AGGREGATE dialog initially show all of the right pane.
The AGGREGATE dialog initially had a very narrow left pane, so
commit 2e6cffa44ef5 "Aggregate dialog: Initialize HPane widget width
on opening" added a callback function to adjust the position of the
divider between the panes to half the window's width. But this
happens after the window has been laid out, which means that unless
the window is very wide to begin with, it will cause part of the
right pane to disappear.
I was able to get more consistently good results by forcing a minimum
width for the left pane with a width_request, and making the right
pane shrinkable below the minimum size needed to show all of it. A
final tweak was to make the left pane automatically grow with the
window.
If you've already used the Aggregate dialog then your preferences for
the window size have been saved, so for testing this it is best to
remove $HOME/.config/psppire before each experiment.
Ben Pfaff [Thu, 12 May 2011 05:36:45 +0000 (22:36 -0700)]
gui: Make File|Recent Files remember the correct encoding.
It's no good to have File|Open... able to open a file in any encoding
if later pulling up the same file with File|Recent Files screws that
up, so this commit fixes the problem by noting the charset in the
file's mime-type.
Ben Pfaff [Fri, 13 May 2011 04:07:12 +0000 (21:07 -0700)]
gui: Recode syntax files on load and save.
This adds a new widget that allows an encoding to be selected to the
File|Open dialog box, as well as to the File|Save As... dialog box
for syntax files only. Files are recoded to UTF-8 on load and back
to the original encoding on save.
File|Recent Files doesn't yet properly save the encoding information.
That's coming up in the next commit.
Bug #33254.
Thanks to John Darrington for reporting this problem.
Ben Pfaff [Thu, 12 May 2011 05:38:57 +0000 (22:38 -0700)]
gui: Refactor checking for .sav and .por suffixes.
It seems reasonable to allow, e.g., .Sav as well as .SAV and .sav, so
this does that. It also splits name_has_suffix() into smaller
pieces that will individually be used in an upcoming commit.
Ben Pfaff [Thu, 12 May 2011 05:27:00 +0000 (22:27 -0700)]
gui: Move null_if_empty_param() from psppire-window to helper.
An upcoming commit will introduce a new user of this function in a
different source file, so it seems best to have this defined in a
common header file.
Ben Pfaff [Thu, 12 May 2011 02:31:22 +0000 (19:31 -0700)]
configure: Use $LIB<x> for linking a program, not $LTLIB<x>.
The gettext manual says:
`@LIBINTL@' is for use without `libtool', `@LTLIBINTL@' is for
use with `libtool'
It has similar wording for libiconv.
Since there's lots of libtool-related stuff in the PSPP makefiles, I
assumed that PSPP needed $LTLIB<x>. But it now appears that this is
only correct for linking a libtool library, not for linking a
program that links against libtool libraries.
On GNU/Linux, both $LIB<x> and $LTLIB<x> expand to the empty string,
so I couldn't tell from experimenting there.
This should fix link failures with Mingw, reported by Harry Thijssen
<pspp@sjpaes.nl>.
Ben Pfaff [Wed, 11 May 2011 05:14:23 +0000 (22:14 -0700)]
pspp-dump-sav: Do not mark any strings for translation.
The pspp-dump-sav program is a debugging tool not intended for common
use by end users, so it seems better to avoid burdening translators
with its strings.
Ben Pfaff [Wed, 11 May 2011 04:35:44 +0000 (21:35 -0700)]
dissect-sysfile: Move to new "utilities" directory as "pspp-dump-sav".
This utility comes in handy from time to time working with end users,
so I think it's best to start installing it with the other binaries,
instead of hiding it in the "tests" directory and having to explain
where to find it.
Ben Pfaff [Tue, 10 May 2011 22:57:44 +0000 (15:57 -0700)]
Automatically link every program against libintl and libiconv.
On glibc systems, libiconv and libintl are both built into libc. That
means that it's very easy to forget to link programs against them,
because it still builds OK on GNU/Linux. However, on other systems
they are usually separate libraries and forgetting to link against
them breaks the build.
This commit switches from previous PSPP practice, in which these
libraries were linked explicitly to every binary that requires them,
to a simplified practice where they are in LIBS so that every binary
is automatically linked to them. This has no disadvantage on
GNU/Linux, our primary target, and the advantage that on other targets
PSPP works more consistently.
Ben Pfaff [Tue, 10 May 2011 05:42:37 +0000 (22:42 -0700)]
gui: Get rid of the_recent_manager global.
The GTK+ function gtk_recent_manager_get_default() is documented to
return a singleton. We might as well take advantage of that instead
of caching its return value and passing it around.
Ben Pfaff [Mon, 9 May 2011 04:52:08 +0000 (21:52 -0700)]
gui: Make File|Recently Used Data behave like File|Open.
File|Open opens a new dataset if the current one is non-empty. Due
to an oversight, File|Recently Used Data always replaced the current
dataset. This commit makes File|Recently Used Data work just like
File|Open.
Ben Pfaff [Mon, 9 May 2011 04:17:51 +0000 (21:17 -0700)]
gui: Make File|Open open multiple datasets by default.
This is what I always intended to make File|Open do, and in fact I
had implemented it that way, but I screwed up in rebasing and forgot
to fix it before my final push.
I'm not 100% certain that this is the correct fix--perhaps the client
is at fault for calling this function before a window exists?--but it
does avoid the message.
Ben Pfaff [Sun, 8 May 2011 05:02:25 +0000 (22:02 -0700)]
gui: Label message output items in the output viewer treeview.
Until now, messages (such as errors and warnings) simply had blank
lines in the output viewer summary treeview. This made for funny gaps
and a generally puzzling appearance. This fixes the problem.
Ben Pfaff [Sat, 7 May 2011 21:19:12 +0000 (14:19 -0700)]
gui: Only unselect range from first data sheet when data changes.
commit 5080068 "gui: Unselect data in data editor when the datasheet
is replaced" unselected all the data in each of the possible four
datasheets (supposing that the "split" view is selected), but if the
"split" view is not in effect then three of the data sheets don't have
windows and this results in "Gdk-CRITICAL **:
gdk_window_invalidate_rect_full: assertion `GDK_IS_WINDOW (window)'
failed".
This commit switches to unselecting the range only from the first data
sheet, which seems to be just as effective at solving the original
problem and does not cause Gdk-CRITICAL messages.
Ben Pfaff [Sat, 7 May 2011 20:01:02 +0000 (13:01 -0700)]
temp-file: Use 64 kB buffer for temporary files.
Generally PSPP uses temporary files because it has a lot of data to
write to them. The default glibc buffer is only 4 kB. In my tests
using a 64 kB reduced runtime from 2.5 seconds to 2.0 seconds, but
increasing the buffer to 1 MB had no additional effect, so 64 kB is a
reasonable choice.
Ben Pfaff [Wed, 4 May 2011 04:30:01 +0000 (21:30 -0700)]
gui: Unselect data in data editor when the datasheet is replaced.
The active cell in the PsppireSheet remembers its contents even if
the whole sheet gets replaced, so we have to unselect the cell when
this happens to avoid replacing the cell's contents by whatever was
there before.
Bug #30502.
Thanks to Michel Boaventura for reporting the problem and John
Darrington for diagnosing it.
Ben Pfaff [Sun, 1 May 2011 05:30:28 +0000 (22:30 -0700)]
gui: Change View|Data and View|Variables to radio button menu items.
I didn't know what View|Variables did until I tried it, at which
point I was surprised that it changed into View|Data. I think that
a pair of radio button menu items better communicates the intent.
Also, this situation seems to fit well what the GNOME Human Interface
Guidelines say about radio button menu items:
Radio button menu items show which of two or more
mutually-exclusive settings are currently selected, and allow
the user to choose a different setting by selecting its menu
item.
* If you need to offer a choice of two mutually-exclusive
settings to the user, use a group of two radio button items
instead of a single check box menu item if the settings are
not clearly opposites. For example, represent View as Icons
and View as List as two radio button items.
Ben Pfaff [Sun, 1 May 2011 05:00:33 +0000 (22:00 -0700)]
gui: Add ellipsis to menu items that require further input.
The GNOME Human Interface Guidelines say:
Label the menu item with a trailing ellipsis ("...") only if
the command requires further input from the user before it
can be performed. Do not add an ellipsis to items that only
present a confirmation dialog (such as Delete), or that do
not require further input (such as Properties, Preferences or
About).
Most PSPPIRE menu items fit this description, so they should have
trailing ellipses, which this commit adds.
Some GTK+ programs use the Unicode HORIZONTAL ELLIPSIS (U+2026)
character instead of three periods in menu items. In a quick check
on my own machine, however, three periods were over twice as common.
In an informal survey on pspp-dev, the best argument presented in
favor of one or the other was that three periods may be more familiar
for translators and thus cause less confusion in translation.
Ben Pfaff [Sun, 1 May 2011 04:08:01 +0000 (21:08 -0700)]
gui: New "entry-dialog" module for prompting for a text string.
This will be used in an upcoming commit to implement File|Rename
Dataset. It is made general-purpose, rather than specifically
for that purpose, because it is not any harder to do so and might
be useful somewhere else in the future.
Ben Pfaff [Wed, 30 Mar 2011 04:29:13 +0000 (21:29 -0700)]
gui: Make syntax execution functions take a PsppireDataWindow argument.
The argument specifies which dataset the syntax is meant to apply to.
For now, this is obvious, because there is only one dataset, but in
the future there will be more than one.
Ben Pfaff [Wed, 8 Sep 2010 04:40:40 +0000 (21:40 -0700)]
gui: Include only <gtk/gtk.h> to use GTK+.
The GTK+ manual says: "The recommended way of using GTK+ has always been
to only include the toplevel headers gtk.h, gdk.h, gdk-pixbuf.h." We
might as well follow this practice, since it only makes life easier.
Ben Pfaff [Sat, 2 Apr 2011 04:14:49 +0000 (21:14 -0700)]
gui: Use dispose instead of finalize method in PsppireDataWindow.
According to the GObject reference manual, "When dispose ends, the
object should not hold any reference to any other member object."
That is, references should be dropped in dispose, not in finalize.